The UK government has rejected a proposed legal "kill switch" that would let ministers shut down AI systems during an emergency. The Cabinet Office said on Friday that blocking access to AI models in the UK would not necessarily stop them from being developed or misused elsewhere.
Liberal Democrat peer Lord Clement-Jones had proposed an amendment to the Cyber Security and Resilience Bill that would create a statutory mechanism to intervene if an autonomous AI model malfunctioned or posed an immediate danger. He argued the measure would provide a "necessary emergency kill switch" where ministers currently lack clear legal authority to order a shutdown.
The government said AI risks need to be addressed through proportionate and targeted measures rather than a blanket power to switch systems off. It added that AI companies have a responsibility to develop their products safely and invest in security measures to protect the technology. Ministers will continue taking a "long-term, science-led approach" to understanding emerging AI threats, including through the UK's AI Security Institute.
Warnings from inside the industry
The decision comes as technology figures issue stark warnings about advanced AI. Last week, Evan Hubinger, an alignment science lead at Anthropic, said he believed AI could potentially kill humanity within the next decade. "We really do earnestly believe AI could kill all humans," Hubinger wrote on X, adding that he personally placed the probability at more than 10% within the next 10 years.
Hubinger also said AI developers have not yet established a reliable solution for ensuring that highly capable systems remain aligned with human interests. His comments followed the resignation of fellow Anthropic researcher Jacob Coxon, who warned that the race among technology companies to develop self-improving AI systems amounted to "gambling with our lives". Coxon accused Anthropic and OpenAI of failing to act responsibly and said people outside the AI industry were underestimating the technology's capabilities.
Anthropic chief executive Dario Amodei has argued for greater restraint, warning that an AI swarm could potentially become capable of taking control of large parts of the internet within a year. SpaceX chief Elon Musk backed the warning, saying: "Dario is right." OpenAI CEO Sam Altman also said he agreed that the development of frontier AI needed to be paced.
Government response and engagement
The government has sought to show it is taking the risks seriously despite rejecting a statutory shutdown mechanism. PM Andy Burnham has elevated AI safety minister Kanishka Narayan's role by bringing him into the Cabinet. Burnham told MPs that the move had "lifted the level of the conversation about artificial intelligence within Government", while inviting politicians across Parliament to take part in discussions about AI safety.
Meanwhile, King Charles is due to meet senior figures from major AI companies as part of an event focused on "AI for good". The meeting, organised under the Ditchley Foundation, is expected to examine whether clearer limits should be established around the use of AI for developing biological weapons and carrying out cyberattacks against critical infrastructure. Executives from OpenAI and Anthropic are expected to attend, alongside Nvidia chief executive Jensen Huang, Google Deep Mind co-founder Demis Hassabis and other technology and ethics figures.
For government professionals tracking these developments, the AI for Government resource hub covers policy responses and regulatory approaches to emerging AI risks. Policy teams seeking structured grounding in these issues may also find the AI Learning Path for Policy Makers useful when preparing briefings on AI governance.
Why this matters for government professionals
The kill switch debate exposes a core tension: Parliament wants clear intervention powers, while the executive argues that blunt shutdown authority would be ineffective against systems hosted or developed abroad. Civil servants working on cyber resilience, national security, or digital policy should expect continued pressure from MPs for specific statutory powers, even if the current bill does not include them. The AI Security Institute remains the government's primary vehicle for assessing threats, so its technical assessments will likely shape what proportionate intervention looks like in practice - and what legislation eventually follows.
Your membership also unlocks:
